Imported from magnus919/agent-skills (lastfm/SKILL.md). Install upstream with npx skills add magnus919/agent-skills --skill lastfm. Copyright stays with the author (MIT).

Last.fm

Setup

export LASTFM_API_KEY="your_api_key_here"
# For write operations only:
export LASTFM_API_SECRET="your_api_secret"
export LASTFM_SESSION_KEY="your_session_key"

Get an API key at https://www.last.fm/api/account/create (free, requires a Last.fm account).

Essential Commands

User queries — "what am I listening to"

Configurable via LASTFM_USERNAME env var. Defaults to <username> if unset.

lastfm-cli user info <username>                      # Profile + stats
lastfm-cli user recent-tracks <username> --limit 10   # Recent listening
lastfm-cli user top-artists <username> --period 7day  # Weekly top artists
lastfm-cli user top-tracks <username> --period overall # All-time faves
lastfm-cli user loved-tracks <username>               # Loved tracks
lastfm-cli user friends <username>                    # Social graph
lastfm-cli user weekly-charts <username>              # Available chart periods

Music Discovery — "what's similar to X"

lastfm-cli artist similar <artist> --limit 10         # Taste graph neighbors
lastfm-cli artist info <artist>                       # Bio, stats, tags
lastfm-cli artist top-tracks <artist>                 # Their most popular
lastfm-cli track similar <artist> "<track>"           # Track-level similarity
lastfm-cli album info <artist> "<album>"              # Tracklist, metadata
lastfm-cli tag top-artists <tag>                      # Genre browsing

Charts — "what's popular"

lastfm-cli chart top-artists --limit 20               # Global
lastfm-cli geo top-artists "Japan" --limit 20         # Per-country
lastfm-cli geo top-tracks "Germany" --limit 20
lastfm-cli tag top-tracks "electronic" --limit 15     # By genre tag

Search — "find that thing"

lastfm-cli artist search "Radiohead"                  # Find by name
lastfm-cli album search "OK Computer"                 # Find albums
lastfm-cli track search "Karma Police"                # Find tracks

Write operations (require session auth)

lastfm-cli track scrobble "Artist" "Song" --album "Album"
lastfm-cli track now-playing "Artist" "Song"
lastfm-cli track love "Artist" "Song"

Auth flow (one-time setup)

# 1. Get a token
lastfm-cli auth get-token

# 2. User visits: https://www.last.fm/api/auth/?api_key=KEY&token=TOKEN

# 3. Exchange for session key
lastfm-cli auth get-session <token>
# → Set export LASTFM_SESSION_KEY=<key>

Reference Files

• references/auth-flow.md — Full walkthrough for setting up write operation auth (scrobble, love, now-playing). Read this before attempting write operations.

Music Discovery Pipeline — "find me new stuff I'll like"

The core flow for turning liked tracks into recommendations:

1. Get user's top artists/tracks from Last.fm:

   lastfm-cli user top-artists <username> --period 1month --limit 5 --json
   lastfm-cli user top-tracks <username> --period 1month --limit 10 --json
   lastfm-cli user loved-tracks <username> --json

2. For each artist, find similar artists via Last.fm's collaborative filtering:

   lastfm-cli artist similar "<artist>" --limit 5 --json

3. For each track, find similar tracks for more granular recommendations:

   lastfm-cli track similar "<artist>" "<track>" --limit 5 --json

4. Cross-reference against what they've already scrobbled (recent-tracks) to filter out already-heard material.

5. Check against the user's own collection (via Radarr/Sonarr/jellyfin or Spotify library) to see what's already in the library vs genuinely new discovery.

Using with --json for Machine Processing

# Pipe to jq for structured queries
lastfm-cli artist similar "Radiohead" --limit 5 --json | jq '.similarartists.artist[] | {name, match}'

# Get a user's all-time top artists with playcounts
lastfm-cli user top-artists <username> --period overall --limit 3 --json | jq '.topartists.artist[] | {name, playcount}'

# Find most popular tracks by a genre
lastfm-cli tag top-tracks "electronic" --limit 10 --json | jq '.tracks.track[] | {name, artist: .artist.name, listeners}'

Known Gotchas

• API key required for EVERY request. The key is free — get one at the Last.fm API account page.
• Rate limit is ~5 req/sec sustained. The API docs say "be reasonable" — if you're making several calls per second continuously, your account may be suspended. Add small delays in loops.
• User-Agent header matters. The CLI sends lastfm-cli/1.0 (hermes-agent). Last.fm's docs explicitly ask for an identifiable User-Agent.
• XML is the default response format. The CLI requests format=json. Without it, responses come back as XML.
• Artist names can be misspelled. Use --autocorrect flag on artist commands to let Last.fm correct misspellings.
• Write operations need auth setup. scrobble, love, and now-playing require a full auth flow (token → session key). Read-only endpoints don't.
• The collaborative filtering graph is NOT social. artist.getSimilar returns algorithmic similarity based on aggregate listening patterns, not human-curated recommendations.
• Timestamps for scrobbles are UNIX epoch seconds. If omitted, uses current time.
• --from and --to on recent-tracks accept ISO 8601 date strings (e.g., 2026-05-27).
• Period values: overall, 7day, 1month, 3month, 6month, 12month.

When to Reach for This Tool

• User asks "what's [person] listening to lately?"
• User wants music discovery: "find me artists similar to..."
• User wants listening statistics: top artists, tracks, albums by period
• User wants geographic music trends: "what's popular in [country]"
• User wants genre exploration via tags
• User wants to scrobble or love tracks programmatically
• Any question about music metadata, artist info, album tracklists

Available Scripts

Prerequisites

• A free Last.fm API key exported as LASTFM_API_KEY (create one at https://www.last.fm/api/account/create); every request fails without it.
• For write operations only: LASTFM_API_SECRET plus a session key from the one-time auth flow in references/auth-flow.md, exported as LASTFM_SESSION_KEY.
• Python 3 on PATH — invoke the bundled tool as python3 scripts/lastfm-cli ... if it is not executable directly; the SKILL.md examples assume lastfm-cli resolves on PATH.
• Optional: LASTFM_USERNAME to set the default user for user queries; jq when piping --json output for structured processing.

Limitations

• Read-only endpoints need only the API key; scrobble, now-playing, and love require the full token → session-key auth flow and will fail without it.
• Sustained call rates above ~5 req/sec risk account suspension — add small delays in loops.
• Similarity endpoints are algorithmic collaborative filtering over aggregate listening patterns, not social or human-curated recommendations.
• Period values are fixed to overall, 7day, 1month, 3month, 6month, and 12month; arbitrary date ranges are only available via --from/--to on recent-tracks.
• Misspelled artist names return empty or wrong results unless --autocorrect is used.

When not to use

Do not use this skill for music playback control (local players, streaming queues), for non-Last.fm catalog data (MusicBrainz, Spotify metadata), or as a lyrics source — the API covers listening history, metadata, charts, tags, similarity, and scrobbling only.